Code accompanying our NeurIPS 2021 traffic4cast challenge

Overview

Traffic forecasting on traffic movie snippets

This repo contains all code to reproduce our approach to the IARAI Traffic4cast 2021 challenge. In the challenge, traffic data is provided in movie format, i.e. a rasterised map with volume and average speed values evolving over time. The code is based on (and forked from) the code provided by the competition organizers, which can be found here. For further information on the data and the challenge we also refer to the competition Website or GitHub.

Installation and setup

To install the repository and all required packages, run

git clone https://github.com/NinaWie/NeurIPS2021-traffic4cast.git
cd NeurIPS2021-traffic4cast

conda env update -f environment.yaml
conda activate t4c

export PYTHONPATH="$PYTHONPATH:$PWD"

Instructions on installation with GPU support can be found in the yaml file.

To reproduce the results and train or test on the original data, download the data and extract it to the subfolder data/raw.

Test model

Download the weights of our best model here and put it in a new folder named trained_model in the main directory. The path to the checkpoint should now be NeurIPS2021-traffic4cast/trained_models/ckpt_upp_patch_d100.pt.

To create a submission on the test data, run

DEVICE=cpu
DATA_RAW_PATH="data/raw"
STRIDE=10

python baselines/baselines_cli.py --model_str=up_patch --resume_checkpoint='trained_models/ckpt_upp_patch_d100.pt' --radius=50 --stride=$STRIDE --epochs=0 --batch_size=1 --num_workers=0 --data_raw_path=$DATA_RAW_PATH --device=$DEVICE --submit

Notes:

  • For our best submission (score 59.93) a stride of 10 is used. This means that patches are extracted from the test data in a very densely overlapping manner. However, much more patches per sample have to be predicted and the runtime thus increases significantly. We thus recommend to use a stride of 50 for testing (score 60.13 on leaderboard).
  • In our paper, we define d as the side length of each patch. In this codebase we set a radius instead. The best performing model was trained with radius 50 corresponding to d=100.
  • The --submit-flag was added to the arguments to be called whenever a submission should be created.

Train

To train a model from scratch with our approach, run

DEVICE=cpu
DATA_RAW_PATH="data/raw"

python baselines/baselines_cli.py --model_str=up_patch --radius=50 --epochs=1000 --limit=100 --val_limit=10 --batch_size=8 --checkpoint_name='_upp_50_retrained' --num_workers=0 --data_raw_path=$DATA_RAW_PATH --device=$DEVICE

Notes:

  • The model will be saved in a folder called ckpt_upp_50_retrained, as specified with the checkpoint_name argument. The checkpoints will be saved every 50 epochs and whenever a better validation score is achieved (best.pt). Later, training can be resumed (or the model can be tested) by setting --resume_checkpoint='ckpt_upp_50_retrained/best.pt'.
  • No submission will be created after the run. Add the flag --submit in order to create a submission
  • The stride argument is not necessary for training, since it is only relevant for test data. The validation MSE is computed on the patches, not a full city.
  • In order to use our dataset, the number of workers must be set to 0. Otherwise, the random seed will be set such that the same files are loaded for every epoch. This is due to the setup of the PatchT4CDataset, where files are randomly loaded every epoch and then kept in memory.

Reproduce experiments

In our short paper, further experiments comparing model architectures and different strides are shown. To reproduce the experiment on stride values, execute the following steps:

  • Run python baselines/naive_shifted_stats.py to create artifical test data from the city Antwerp
  • Adapt the paths in the script
  • Run python test_script.py
  • Analyse the output csv file results_test_script.csv

For the other experiments, we regularly write training and validation losses to a file results.json during training (file is stored in the same folder as the checkpoints).

Other approaches

  • In naive_shifted_stats we have implemented a naive approach to the temporal challenge, namely using averages of the previous year and adapting the values to 2020 with a simple factor dependent on the shift of the input hour. The statistics however first have to be computed for each city.
  • In the configs file further options were added, for example u_patch which is the normal U-Net with patching, and models from the segmentation_models_pytorch (smp) PyPI package. For the latter, smp must be installed with pip install segmentation_models_pytorch.
Owner
Nina Wiedemann
Nina Wiedemann
Randomized Correspondence Algorithm for Structural Image Editing

===================================== README: Inpainting based PatchMatch ===================================== @Author: Younesse ANDAM @Conta

Younesse 116 Dec 24, 2022
Code For TDEER: An Efficient Translating Decoding Schema for Joint Extraction of Entities and Relations (EMNLP2021)

TDEER (WIP) Code For TDEER: An Efficient Translating Decoding Schema for Joint Extraction of Entities and Relations (EMNLP2021) Overview TDEER is an e

Alipay 6 Dec 17, 2022
Python project to take sound as input and output as RGB + Brightness values suitable for DMX

sound-to-light Python project to take sound as input and output as RGB + Brightness values suitable for DMX Current goals: Get one pixel working: Vary

Bobby Cox 1 Nov 17, 2021
Seeing All the Angles: Learning Multiview Manipulation Policies for Contact-Rich Tasks from Demonstrations

Seeing All the Angles: Learning Multiview Manipulation Policies for Contact-Rich Tasks from Demonstrations Trevor Ablett, Daniel (Yifan) Zhai, Jonatha

STARS Laboratory 3 Feb 01, 2022
We will release the code of "ConTNet: Why not use convolution and transformer at the same time?" in this repo

ConTNet Introduction ConTNet (Convlution-Tranformer Network) is proposed mainly in response to the following two issues: (1) ConvNets lack a large rec

93 Nov 08, 2022
Discovering Explanatory Sentences in Legal Case Decisions Using Pre-trained Language Models.

Statutory Interpretation Data Set This repository contains the data set created for the following research papers: Savelka, Jaromir, and Kevin D. Ashl

17 Dec 23, 2022
Neural Fixed-Point Acceleration for Convex Optimization

Licensing The majority of neural-scs is licensed under the CC BY-NC 4.0 License, however, portions of the project are available under separate license

Facebook Research 27 Oct 06, 2022
Repository for MuSiQue: Multi-hop Questions via Single-hop Question Composition

🎵 MuSiQue: Multi-hop Questions via Single-hop Question Composition This is the repository for our paper "MuSiQue: Multi-hop Questions via Single-hop

21 Jan 02, 2023
Open source Python module for computer vision

About PCV PCV is a pure Python library for computer vision based on the book "Programming Computer Vision with Python" by Jan Erik Solem. More details

Jan Erik Solem 1.9k Jan 06, 2023
RealFormer-Pytorch Implementation of RealFormer using pytorch

RealFormer-Pytorch Implementation of RealFormer using pytorch. Includes comparison with classical Transformer on image classification task (ViT) wrt C

Simo Ryu 90 Dec 08, 2022
Face and Pose detector that emits MQTT events when a face or human body is detected and not detected.

Face Detect MQTT Face or Pose detector that emits MQTT events when a face or human body is detected and not detected. I built this as an alternative t

Jacob Morris 38 Oct 21, 2022
Tensorflow Repo for "DeepGCNs: Can GCNs Go as Deep as CNNs?"

DeepGCNs: Can GCNs Go as Deep as CNNs? In this work, we present new ways to successfully train very deep GCNs. We borrow concepts from CNNs, mainly re

Guohao Li 612 Nov 15, 2022
Applying curriculum to meta-learning for few shot classification

Curriculum Meta-Learning for Few-shot Classification We propose an adaptation of the curriculum training framework, applicable to state-of-the-art met

Stergiadis Manos 3 Oct 25, 2022
Image Deblurring using Generative Adversarial Networks

DeblurGAN arXiv Paper Version Pytorch implementation of the paper DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks. Our netwo

Orest Kupyn 2.2k Jan 01, 2023
The second project in Python course on FCC

Assignment Write a function named add_time that takes in two required parameters and one optional parameter: a start time in the 12-hour clock format

Denise T 1 Dec 13, 2021
Exploit ILP to learn symmetry breaking constraints of ASP programs.

ILP Symmetry Breaking Overview This project aims to exploit inductive logic programming to lift symmetry breaking constraints of ASP programs. Given a

Research Group Production Systems 1 Apr 13, 2022
A tutorial on DataFrames.jl prepared for JuliaCon2021

JuliaCon2021 DataFrames.jl Tutorial This is a tutorial on DataFrames.jl prepared for JuliaCon2021. A video recording of the tutorial is available here

Bogumił Kamiński 106 Jan 09, 2023
Deep Crop Rotation

Deep Crop Rotation Paper (to come very soon!) We propose a deep learning approach to modelling both inter- and intra-annual patterns for parcel classi

Félix Quinton 5 Sep 23, 2022
This is the official pytorch implementation for the paper: Instance Similarity Learning for Unsupervised Feature Representation.

ISL This is the official pytorch implementation for the paper: Instance Similarity Learning for Unsupervised Feature Representation, which is accepted

19 May 04, 2022
Code for EMNLP 2021 paper: "Learning Implicit Sentiment in Aspect-based Sentiment Analysis with Supervised Contrastive Pre-Training"

SCAPT-ABSA Code for EMNLP2021 paper: "Learning Implicit Sentiment in Aspect-based Sentiment Analysis with Supervised Contrastive Pre-Training" Overvie

Zhengyan Li 66 Dec 04, 2022